Talk at South-by-southwest (SXSW) on crowdsourcing + human computation = crowd computing., March 11, 2016. See http://www.humancomputation.com.
Global growth in Internet connectivity and participation is driving a renaissance in human computation: use of people rather than machines to perform certain computations for which human competency continues to exceed that of state-of-the-art algorithms (e.g. “AI-hard” tasks such as interpreting text or images). While current AI limitations will certainly improve with time, using human computation lets us bulid applications which deliver superior results today. Just as cloud computing now enables us to harness vast Internet computing resources on demand, new crowdsourcing APIsenable us to build computing systems which integrate human computation at run-time, invoking crowd labor on-demand and at-scale. Moreover, we can achieve the best of both worlds by integrating automated AI with human computation, creating hybrid systems with capabilities greater than the sum of their parts. When AI falls short, not only can human computation meet the immediate end-user need, but the results can be fed back into the system to further improve the AI. As a consequence, AI limitations are no longer a bottleneck to delivering innovative, new applications. Such enhanced capabilities have begun to change how we design and implement intelligent systems. While early work in crowd computing focused only on collecting more data from crowds to better train AI, we are increasingly seeing hybrid, socio-computational system emerge which creatively blend human computation and AI at run-time to solve hard computing problems. As such, we find ourselves today in an exhilarating new design space in which intelligent system capabilities are seemingly limited only by our imagination and creativity in designing new algorithms to compute effectively using crowds as well as silicon.